First informal meeting about Virtualisation with Xen

Tuesday 29 May 2007, 16:00 → 17:00 Europe/Zurich

31/1-012 (CERN)

Sharing ideas and plans about Xen Virtualisation at CERN, between FIO, OpenLab, GD, with inputs from Netops

Dear all,

As decided at our first meeting last 29th of May, we will meet again next Tuesday 12th of June 4pm same place.

We will focuss on Network aspects of Virtualisation.

Luna will present us a summary of his survey on the subject.

A part from that:

- action list of last meeting was:

-----------------

-o- network aspects to be further discussed (switch boundaries, delays for DNS updates/propagation,…)

-o- Quattorisation of Xen boxes of Andreas U. as an exercise/prototype

-o- VO images ? What is the request ? Should VO's be able/allowed to install their own images on WN's ? At CERN ? Outside CERN ? What about access to VM ? Root access ?

- notes:

-----------

Present were: Luna, Alex, Sverre, Havart, Xavier, Andreas H., Veronique, Andreas U.

Openlab; Hplab works on virtualisation for gLite testbed, look into migration possibilities. Also investigates about running VM on WN's (in collaboration with Louis Poncet).

"OSFarm": creation of images from SLC, Debian, (from GD), management of user images.

Timescales: Openlab: development phase. Started last summer. More progress expected by next summer. Summer students will join.

NetOps: Andreas H. is waiting CS to be available for discussions. A design is ready to be presented to them, for large number of VM's, for gateways with GPN, …

Suggestions/ideas:

-Suggestion to use LSF for load-balancing of hypervisors, for WN's

- CE,SE,… need to be migratable anywhere in the CC (note: we currently have 22 CE's at CERN)

- use VM to run WN on different OS's

- will VO send their own images ?

- currently VO's install their own sw in their own directories on WN's, why not their own image ?

- 15 minutes are needed currently for an IP change, + time for DNS propagation

- for hardware availability, having services running on VM's would ease dynamic access to hw. , better scheduling for draining

- what about ETICS ?